Lightning-AI/pytorch-lightning · error · TypeError

The FSDP strategy can only work with the `FSDPPrecision` plu

Error message

The FSDP strategy can only work with the `FSDPPrecision` plugin, found {precision}

What it means

FSDP requires mixed-precision handling to go through the FSDPPrecision plugin (which configures torch's MixedPrecision policy). The precision setter raises TypeError when given any other Precision implementation.

Source

Thrown at src/lightning/fabric/strategies/fsdp.py:247

        plugin = self.precision
        if isinstance(plugin, FSDPPrecision):
            return plugin.mixed_precision_config
        return None

    @property
    @override
    def precision(self) -> FSDPPrecision:
        plugin = self._precision
        if plugin is not None:
            assert isinstance(plugin, FSDPPrecision)
            return plugin
        return FSDPPrecision("32-true")

    @precision.setter
    @override
    def precision(self, precision: Optional[Precision]) -> None:
        if precision is not None and not isinstance(precision, FSDPPrecision):
            raise TypeError(f"The FSDP strategy can only work with the `FSDPPrecision` plugin, found {precision}")
        self._precision = precision

    @override
    def _configure_launcher(self) -> None:
        assert self.cluster_environment is not None
        if not self.cluster_environment.creates_processes_externally:
            self._launcher = _SubprocessScriptLauncher(self.cluster_environment, self.num_processes, self.num_nodes)

    @override
    def setup_environment(self) -> None:
        super().setup_environment()
        self._setup_distributed()

        # if 'device_mesh' in the `_fsdp_kwargs` is provided as a tuple, update it into the `DeviceMesh` object here
        if isinstance(self._fsdp_kwargs.get("device_mesh"), tuple):
            from torch.distributed.device_mesh import init_device_mesh

            self._fsdp_kwargs["device_mesh"] = init_device_mesh("cuda", self._fsdp_kwargs["device_mesh"])

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Pass a precision string and let the strategy create FSDPPrecision, e.g. FSDPStrategy() with Fabric(precision="16-mixed")
2. If constructing manually, instantiate FSDPPrecision (e.g. FSDPPrecision("16-mixed"))
3. Ensure any strategy-setup hook does not overwrite precision with a foreign plugin

Example fix

# before
strategy = FSDPStrategy(precision=SomeCustomPrecision())

# after
from lightning.fabric.strategies.fsdp import FSDPStrategy
fabric = Fabric(precision="16-mixed", strategy=FSDPStrategy())

from lightning.fabric.plugins.precision.fsdp import FSDPPrecision
assert precision is None or isinstance(precision, FSDPPrecision)
strategy = FSDPStrategy(precision=precision)

from lightning.fabric.plugins.precision.fsdp import FSDPPrecision
def valid_fsdp_precision(p) -> bool:
    return p is None or isinstance(p, FSDPPrecision)

When it happens

Trigger: FSDPStrategy(precision=Precision()) with a custom subclass, or assigning strategy.precision = DeepSpeedPrecision()/MixedPrecisionPlugin/... at runtime; also fabric code that injects a non-FSDPPrecision plugin before setup.

Common situations: Sharing a precision plugin across strategies in a config dict; upgrading Lightning versions where precision plugins changed classes; constructing FSDPStrategy with precision="16-mixed" via a path that creates a non-FSDP plugin.

Related errors

• `precision={precision!r})` is not supported in FSDP. `precis
• `precision={precision!r}` does not use a scaler, found {scal
• Gradient clipping with FSDP is only possible if the module p
• `activation_checkpointing_policy` must be a set, found {poli
• Blocking backward sync is only possible if the module passed

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/2732e0dcefb987d6. Report an issue: GitHub.